Home / Learn / Data & AI / Offline AI / Lesson 16.4
All lessons

Module 16: Cost: local AI against subscriptions · Lesson 16.4 · about 20 min

Break-even: when local becomes cheaper

Work out how many months it takes for a local machine to cost less than the plan or API spend it replaces, for light, medium and heavy users, with a calculator in your own currency.

Key points, meanings and formulas on one page: choose "Save as PDF" in the print box.

Module 16 · about 20 minutes · read first: 16.1 and 16.3. You’ll be able to work out, in your currency, when a local machine pays for itself.

A local machine pays for itself when the money you stop spending each month has added up to what the machine really cost you.

1. Why this matters

Everyone asks the same question: “Will it pay for itself?” The answer depends on three numbers: what you spend on cloud AI now, what the hardware really costs you (Lesson 16.3), and what it costs to run. This lesson turns them into a number of months, called the break-even or payback period.

2. The calculator

Pick your country; the plan, hardware and electricity price fill in from this course’s price file, and you can type your own.

Calculator: break-even

Local (hardware, then electricity) Cloud plan

Hardware prices checked 18 November 2026, plan prices 19 November 2026 (sources in Lessons 5.1 to 5.4 and 16.1). A price shown in US$ means no local price was found: type your own. Electricity counts 30 days a month. Leave out the hardware price if you would buy the computer anyway, and count only the extra (for example the larger memory). This ignores your time, repairs and the fact that a local model may not match the cloud model's quality (Lesson 17.1).

The solid line is local: it starts at the hardware price (after resale) and climbs slowly with electricity. The dashed line is the plan: it starts at zero and climbs every month. Where they cross is the break-even.

What they spend now. A 6-person team in Toronto pays for six US$20 plans: about CA$190 a month with HST and exchange fees.

What they buy. A 64 GB Mac mini for the office at about CA$3,800. They expect to sell it for 40% later, so its real cost is 3,800 × 0.6 = CA$2,280.

What they still pay. They keep two plans for the hardest jobs (about CA$63) and the machine adds about CA$3 of power. Monthly saving: 190 − 63 − 3 = CA$124.

Divide. 2,280 ÷ 124 ≈ 18 months. After that, every month saves CA$124, and client files stay in the office from day one.

months=H−SC−R
H − S
what the machine really costs: price minus what you expect to sell it for
C
the cloud spend you stop each month (plans cancelled or API bill avoided)
R
what running locally costs each month: power, plus any plans you keep
C − R
the saving each month; if it is zero or less, the machine never pays back
Physical meaning
It is a race between a head start and a speed. The cloud starts at zero but climbs fast; local starts high but climbs slowly. Break-even is the month the slow climber is caught, and after it every month is money kept.

3. Three typical users

Made-up usage, US list prices.

User Cloud spend now Local choice Break-even
Light: a few chats a day Free tier, or Google AI Plus at US$4.99 Existing laptop, small model Nothing to buy: free from day one, but the plan was cheap too
Medium: daily writing and coding One US$20 plan Mac mini 32 GB, counting only the US$400 extra About 20 months on the full US$400, about 12 counting 40% resale; power is pennies
Heavy: agents and long jobs all day US$200 plan or API spend 128 GB AI box at US$3,999 to US$4,699 20 to 26 months; faster if the API bill was higher

Two lessons from the table. Medium users rarely save much money by replacing one US$20 plan; they go local for privacy, offline use or no limits. Heavy users and teams are where local pays back clearly, especially when one machine replaces several seats (Lesson 5.5).

Maths and principle

Principle: break-even is when the money saved each month has paid back the money spent up front.

Equation: months to break even = (hardware − resale) ÷ (cloud spend per month − local running cost per month). If the bottom is zero or less, it never breaks even.

Analogy: solar panels. They cost a lot up front and save a little each month; the question is how many months until the savings catch up.

Try it: a ₹1,50,000 machine, 30% resale, replacing ₹10,699 a month, ₹500 a month electricity? ((1,50,000 × 0.7) ÷ (10,699 − 500) = 1,05,000 ÷ 10,199 ≈ 10 months.)

“If it pays back on paper, going local is the right call.” Money is only half the comparison. If the local model does the job worse than the cloud model you dropped, the extra checking and fixing can eat the saving. Run the quality test in Lesson 17.1 before you trust the break-even.

4. Worked example

Made-up team, real list prices. A 6-person team in Toronto pays for six US$20 plans (US$120 a month, plus HST and exchange fees, about CA$190). They buy a 64 GB Mac mini for the office at about CA$3,800 and keep two plans for the hardest jobs (about CA$63 a month). Saving: about CA$127 a month, minus about CA$3 of electricity. Break-even: 3,800 × 0.6 (after 40% resale) ÷ 124 ≈ 18 months. They also gain one place where client files stay in the office.

5. Practice

  1. Run the calculator with your own plan and the machine you would buy.
  2. What happens to break-even if you keep one cloud plan instead of cancelling all?
  3. Find the electricity price at which the RTX 5090 option never breaks even against a US$20 plan, at 3 hours a day.
Answers
  1. Your answer; note the date and source of each price.
  2. The monthly saving shrinks, so break-even comes later, but you keep the best cloud model for hard jobs.
  3. 750 W × 3 h × 30 days = 67.5 kWh a month; 20 ÷ 67.5 ≈ US$0.30 per kWh. Above that, it never pays back on money alone.

Try question 3 with other machines and hours:

New words: break-even, payback period, seat (one person’s plan).

  • Break-even months = (hardware − resale) ÷ (cloud spend stopped − local running cost each month).
  • If running costs match or beat the cloud spend, the machine never pays back on money alone.
  • One US$20 plan rarely pays back fast; heavy users and teams replacing several seats pay back in one to two years.
  • Keeping one cloud plan for hard jobs slows break-even but keeps the strongest model on hand.
  • Check quality too: a cheaper answer that needs fixing is not cheaper.

Check yourself

1When does local never become cheaper on money alone?
2A heavy user cancels a US$200 plan and buys a US$4,000 box. Roughly when does it pay back, ignoring power and resale?
3Why might the break-even answer mislead you even if the maths is right?

Sources

  • Hardware prices: prices.ts, checked 18 November 2026 (Lessons 5.1 to 5.4)
  • Plan prices: prices.ts, checked 19 November 2026 (Lesson 16.1 sources)
  • Home electricity prices: Lesson 6.2 sources
  • Brealey, R. A., Myers, S. C. and Allen, F., Principles of Corporate Finance, McGraw-Hill (the payback period, in the chapter on investment decision rules)
  • U.S. Energy Information Administration, Electric Power Monthly (average retail electricity prices, official page), https://www.eia.gov/electricity/monthly/

Your ticks stay in this browser. Save them as a file on the course home.

Updated 2026-11-22. Prices and model names change fast: check the official source before you buy.